Predicting and Parsing Language in Time and Space
Committee Chair: Mariliina Pylkkanen
Committee Members: Alec Marantz, Chris Barker, Brian McElree, Colin Phillips (UMD).
We use our experience to navigate the world in the present, and to predict what might come up in the near future. In this dissertation, I ask how prediction might help explain the rapidity and efficiency of language processing. In four experiments, I show, for the first time, that brain regions dedicated to low-level sensory analysis are sensitive to seemingly high-level factors of language processing as early as 100 ms after the presentation of a word. To explain this finding, I propose a Sensory Hypothesis: From contextual predictions, the brain derives form-feature estimates for syntactic categories or lexical-semantic representations, and these estimates are made available to sensory cortices by way of top-down modulation. A mismatch with a form expectation then causes enhanced neural activity in sensory regions.
While many scholars accept that prediction plays a crucial role in language processing, the neural correlates of linguistic prediction itself, or the preactivation of predicted linguistic representations, have not been previously investigated. I here compare highly predictive contexts to cases where no expectations are generated and show that left-temporal cortex (preactivation of lexical representations) and visual cortex (preactivation of form features) show increased brain activity for highly predictive environments right before the expected word is presented. The same brain areas were more active for words that violated lexical-semantic expectations than for words that satisfied predictions, but in reverse temporal order. As such, this research takes an important step towards elucidating the mechanisms by which prediction allows rapid language processing.